Research Scientist, Memory, Reasoning and Continual Learning, Deepmind

Google Google · Big Tech · Toronto, ON +1

Research Scientist at Google DeepMind focused on advancing AI in Memory, Reasoning, and Continual Learning. The role involves initiating novel research, designing and executing experiments on topics like RAG, continual learning, and multi-step reasoning, developing evaluations for complex capabilities, and building infrastructure for advanced AI systems. The position requires a PhD and experience in AI research, with a focus on publishing findings and contributing to the research community.

What you'd actually do

  1. Initiate and lead novel research directions, applying insights from machine learning and related fields (e.g., NLP, Reinforcement Learning (RL), computational neuroscience) to advance long-context attention mechanisms, Retrieval-Augmented Generation (RAG), continual learning architectures, and multi-step reasoning frameworks.
  2. Design and execute end-to-end experiments, proposing and testing hypotheses to better align model-based agents, mitigate catastrophic forgetting, and improve sample efficiency in non-stationary environments.
  3. Develop evaluations and benchmarks that stress-test long-horizon memory, out-of-distribution generalization, and complex planning capabilities, including in-depth search debugging of failure modes.
  4. Build and improve infrastructure for high-capacity context windows, dynamic memory structures, and continuous training pipelines in close collaboration with engineering teams.
  5. Communicate research findings clearly through plots, writeups, and paper-ready narratives, while contributing to a team culture of first-principles thinking, high standards, and constructive feedback.

Skills

Required

  • PhD in Computer Science, Machine Learning, Mathematics, Cognitive Science, or a related technical field, or equivalent practical experience.
  • 2 years of experience in artificial intelligence research, including publications in conferences or journals (e.g., NeurIPS, ICML, ICLR).
  • Experience with Deep Learning, Reinforcement Learning, Natural Language Processing, or architectures for Continual/Lifelong Learning.
  • Experience in Python and deep learning framework (e.g., JAX, TensorFlow, PyTorch).
  • Experience in algorithms design, running experiments, and analyzing results.

Nice to have

  • Postdoctoral or equivalent industry research experience focusing on large language models, autonomous agents, or long-context memory systems.
  • Experience designing novel benchmarks or evaluation frameworks for complex planning and reasoning capabilities.
  • Knowledge of the interdisciplinary intersection between machine learning and neurobiology or computational neuroscience.
  • Ability to contribute to open-source ML software or experience in collaborating with cross-functional teams.
  • Familiarity with distributed training techniques and building infrastructure for high-capacity context windows.

What the JD emphasized

  • publications in conferences or journals (e.g., NeurIPS, ICML, ICLR)
  • novel research directions
  • Retrieval-Augmented Generation (RAG)
  • continual learning architectures
  • multi-step reasoning frameworks
  • model-based agents
  • catastrophic forgetting
  • sample efficiency
  • long-horizon memory
  • out-of-distribution generalization
  • complex planning capabilities
  • high-capacity context windows
  • dynamic memory structures
  • continuous training pipelines

Other signals

  • novel research directions
  • advance long-context attention mechanisms
  • Retrieval-Augmented Generation (RAG)
  • continual learning architectures
  • multi-step reasoning frameworks
  • model-based agents
  • catastrophic forgetting
  • sample efficiency
  • long-horizon memory
  • out-of-distribution generalization
  • complex planning capabilities
  • high-capacity context windows
  • dynamic memory structures
  • continuous training pipelines